Analysis and Testing of Mobile Wireless Networks

Richard Alena, Darin Evenson*, Victor Rundquist'_NASA Ames Research Center

Moffett Field, CA 94035-1000

(650) 604-0262ralena@mail.arc.nasa, gov

* Naval Postgraduate School, daevenso@nps.navy.mil

t DeAnza College, vmndquist@mail.arc.nasa.gov

Abstract-Wireless networks are being used to connect

mobile computing elements in more applications as the

technology matures. There are now many products (suchas 802.11 and 802.11b) which run in the ISM frequency '

band and comply with wireless network standards. Theyare being used increasingly to link mobile Intranet intowired networks. Standard methods of analyzing and

testing their performance and compatibility are needed todetermine the limits of the technology.

This paper presents analytical and experimental methodsof determining network throughput, _mge and coverage,_ad in_i-fe_ce--s_ces_--Both radio frequency OLF)

domain and network domain analysis have been appliedto determine wireless network throughput and range in the

outdoor environment. Comparison of field test data taken

under optimal conditions, with performance predictedfrom RF analysis, yielded quantitative results applicable

to future designs.

TABLE OF CONTENTS

AbstractIntroduction

Wireless Network Technologies

Mobile Exploration System OverviewWireless Network Metrics

RF Propagation TheoryMEX Testbed Data

RF Link AnalysisInterference And ScalabilityConclusions

References

Layering multiplewireless network subnets can increaseperformance. Wtreless network components can be set todifferent radio frequency-hopping sequences or spreading

functions, allowing more than one ,,ubnet to coexist.Therefore, we ran multiple 802.11-compliant systemsconcatrrently in the same geographical area to determineinterference effects and scalability. The results can be

used to design of more robust networks which have

multiple layers of wireless data communication paths and

provide increased throughput overall.

Copyright 2002 by the Institute of Electrical and Electronics Engineers.

No copyright is asserted in the United States under Title 17, U.S. Code,

The U. S. Government has royalty-free license to exercise all rights

under the copyright claimed herein for governmellt purposes. All other

fights are reserved bythe copyright owner.

Paper 410.

https://ntrs.nasa.gov/search.jsp?R=20020088428 2018-07-05T03:58:17+00:00Z

INTRODUCTION

By nature, wireless networks differ from wired networks

by the transient, ad-hoc connections they establisk Theterm wireless local area network (WLAN) represents acommunications method or system established through

the use of radio frequency (RF) technology that can func-

tion either as an extension to an existing local area net-work (LAIN-) or as an alternative for a wired LAN.

Relying on complex spread-spectrum radio signals, thebandwidth and connection quality ,_ary dramatically de-

pendant upon free space propagation characteristics. Thenetworks are increasingly used to connect mobile com-

puting elements in indoor and outdoor environments, cov-ering distances ranging from sever:d meters to several

kilometers and providing bandwidths from hundreds ofkilobits per second to several megabils per second. [1]

The Intelligent Mobile Technologies (/MT) team at

NASA Ames Research Center studied the integration of

disparate collections of mobile nodes consisting of sen-sors, notebook computers, servers, gzaplfics tablets, and

display devices using both 802.11 and 802.1 lb wirelessnetwork products. [2] The team developed a pragmatic ap-

proach to testing and defining the ch_u'acterisfics of theseproducts consisting of mathematical modeling, network

performance metrics, radio frequency (PF) domain analy-- sis and field testing. [3] The metrics show sustainable and

peak network throughput, range and coverage resultswhich can be correlated with interference effects. Related

to interference is scalability, where non-interfering sys-

tems can be layered to provide a much greater aggregatebandwidth capability.

ications, more bandwidth is used than in conventionalnarrowband transmission that is based on a specific radio

frequency. This results in an inefficient use of bandwidth,

but it mitigates the risks of staying on one frequency andbeing intercepted or jammed. Because the signal is spreadover a greater frequency spectrum, the receiver must

know the parameters of the original signal in order to re-cover it correctly. If the receiver is not tuned to the correct

frequency at the correct time, the signal will appear as

background noise.

With frequency-hopping spread-spectrum (FHSS) com-munications, a narrowband carrier radio frequency is

shifted in discrete increments of frequency, predetermined

by a code. The amplitude, and thus the power and energyin each hop are constant. The frequency shift is based on a

pattern generated from a code called the hopping se-quence that spreads transmission over a wide frequencyband. For the signal to be received correctly, the hopping

sequence must be known in advance by both the receiverand the transmitter. With direct-sequence spread-spectrum

commurficafions (DSSS), a narrowband carrier radio fre-

quency is modulated by a digital "chipping code" with thecode bit rate being larger than the information bit rate thusspreading the signal over a very" large bandwidth. Because

the "chipping code" adds redundancy to the information

being transmitted, this permits a receiver to recover theoriginal data even ff one or more bits are damaged duringtransmission. Because the DSSS technique spreads infor-marion over a wide bandwidth, a receiver without the cor-

rect "chipping code" views the signal as low power wide-band noise.

This paper presents a brief overview of WLAN technolo-

gies, analysis methods, performance testing and field testresults. An introduction to RF propagation theory and asimple model of the access protocol provide some inter-

esting background for interpreting the results. The goal ofthis effort is to provide performance data relevant to thedesign and application of next-generation WLAN

systems.

WIRELESS NETWORK TECHNOLOGI£S

There are several technologies that are used to createWLANs, including frequency-hopping spread-spectnun(FHSS) and direct-sequence spread-specmma (DSSS)

communications. A discussion of spread-spectnnn tech-nologies is important for understanding the characteristics

of WLANs and the difference in test results.[4] Spread-

spectrum communications were developed during the late1940's as a mechanism to provide a reliable and securecommunications method for the military under combat

conditions. [5] Spread-spectrum is the major technolog3 _ofchoice in the commercial WLAN induslry because manydevices can be used simultaneously wilh minor interfer-

ence concerns. With the use of spread-spectrum commun-

Compliance with standards ensures appliance inter-

operability with other WLAN components from differentvendors that have adopted the same standard. [6] In theUnited States, the FCC governs radio transmissions, andmost WLANs broadcast over one of the Insmnnentation,

Scientific, and Medical (ISM) bands. These include 902-

928 MHz, 2.4-2.483 GHz, 5.15-5.35 GHz, and 5.725-5.875 GHz. In 1997, when the first internationally recog-nized IEEE 802.11 standard arrived, there was a sudden

surge of market interest, resulting in the advent of higher-speeds with 802.11b and 802.11a. The 802.11 standardrefers to the media access control (MAC) layer of the ISO

network model and also defines modulation type and data

rotes. Quickly summarizing, 802.11 defines a 2 Mbit/secdata rate, and 802.11b defines an 11 Mbit/sec data rate in

the ISM frequency band. This study tested both types of

WLAN systems.

In a typical WLAN configuration, a transmitter andreceiver, or transceiver device, called an access point

(AP), connects to the wired network from a fixed locationusing standard cabling. At a minimum, the AP receives,buffers and transmits data between the WLAN and the

wired network infrastructure. A single ALP can support a

smallgroupofusersandfimctionswithinarangeoflessthanonehundredto severalhundredfeet.Theaccesspointmaybemountedanywhere_tt ispracticalaslongas the desiredradiocoverageis obtained.TheAPprovidesa stoneceil or subnetof WLANcoverage.MultipleAPsprovidemultiplesubnets,allowingmobileusersto roamfromonesubnetto anothermaintainingconnectionto thenetwork,muchlike a cellularphonesystem.

To solveparticularproblemsof WLAN topology, thenetwork designer might choose to use an extension point

(EP) to augment the network of APs. EPs (or repeaters)extend the range of the network by relaying radio signalsfrom an ALP tO a station or another EP. Extension points

may be strung together in order to pass along messagingfrom an AP to far-fhmg clients. One last item of WLAN

equipment to consider is the directional antenna.Directional antennas increase the gain, or relative focus,of the antenna which increases the range and the

reliability of a given signal. The other option to increase

range would be to substantially increase power output, butthis is not always desirable in a military environment

where tactical operators are focused on clandestineoperations. It is also a concern in mobile apphcations,where minimum power consumption is desired.

MOBILE EXPLORATION S YSTEM OVERVIEW

The Mobile Exploration System (M_::X) is a testbed fordeveloping communication system architecture for

planetary exploration. The MEX is a project of the

Human Exploration and Development of Space (HEDS)Enterprise at NASA.[7] The testbed is used for technol-

ogy development of computational components, com-mtmications _stems, and collaborative applications. TheMEX is used for developing operations scenarios for

planetary exploration and is being tested through a seriesof field trials. Recently the MEX was taken to the NASA

Haughton Mars Project (I-IMP), a Mars analog field re-search site located in the Canadian High Arctic, an ex-

tremely remote and hostile environment, one that waschosen for its similarity to the planet Mars. [8] The MEXis assembled from commercial technologies, and com-bines characteristics of both satellite and surface com-

munication systems. Satellite communications are for the

long haul connections between major nodes, such as the

planetary Base to Mission Control on earth.[9] Theplanetary surface communications are designed for highbandwidth, low latency connections for operations con-

ducted for exploration. The nodes are compact and low

power, and support many special purpose nodes with an-cillary functions such as position estimation, imaging,data collection and environmental monitoring. The sur-

face systems of the MEX are the focus of the tests con-

ducted for this paper. [10]

The MEX communication systems include Proxim's

RangeLAN2 radios for WLAN implementation. A second

WLAN from the Orinoco product line was incorporatedinto the MEX architecture this year, representing an

802.11b product. The computers all ran MicrosoftWindows operating systems, and all communicationsoccurred in the unlicensed 2.4 Ghz ISM band.

..... ....... _,.--,_::,...... .... _:,_-,. -I t3SatCom:

......... ! I. : _:- . . ..:. f_ :.

pr*ximi:R_'-I_N. "_ki: IRO_ "_

vin,* camQra:_:_t:/_ !G._:hiCS

Hand:h e:id::in:strurnents t:_

[:._4<-i :

Figure 1. Mobile exploration system, 2001

Theprimarygoalwasto characterizeandcomparetheperformanceofthetwoWLANsin fieldconditions.Bothproductsareactuallydesignedfor indoorshort-rangecommunicationsand were adaptedto the outdoorenvironmentandlongerrangecommunicationsthroughthe useof environmentalenclosuresand high-gaindirectionalantennas.An overviewof the MEXarchitectureandits individualcomponentsrangingfromhandhelddevicesandsmallsensors,toruggedizedserversmountedonanall terrainvehicle(ATV)is showninFigure1.

The WLAN componentsonginallytestedare fromProxim,Inc.,whichwasthefirst-to-marketin 1994withits RangeLAN2productsoperatingin the 2.4 Ghz fre-quency band. At that time, Proxim's OpenAir protocolwas proprietary and there was no WLAN standard.

Proxim was part of the committee specifying the proposedIEEE 802.11 protocol, and soon after the release of thestandard, Proxim adopted it. However they also had a

huge market share based on the OpenAir protocol and de-cided to support both product lines. The RangeLAN2

WLAN components tested operate according to the Open-Air specification at data rates up to 1.6 Mbps, but, given

their historical contribution to the 802.11 protocol, are

directly comparable to 802.11 products operating at 2.0Mbps. Figure 2 shows some of the testbed components inthe field.

WIRELES S NETWORK METRIC S

The physical characteristics that influence WLAN per-

formance can be measured using a few key parameters.

"Range" is the distance over which a WLAN can com-municate and is a function of transmitted power and

receiver sensitivity, antennas used and propagation loss.Interactions with typical building objects, including walls,

metal and even people, can affect how energy propagatesand thus what range a particular system achieves. The

range for typical WLAN systems varies from under 100feet to more than 20 miles. "Coverage" is the geometrical

distribution of link viability and is given as a set of anglesin three dimensions and is generally determined by an-

tenna choice. Range can be extended by increasing power

or by using directional antennas or repeaters.

"Throughput" is the sustainable data rate through a

WLAN given in bits per second Cops) or bytes per second.Peak and raw data rates are related measures. Factors that

affect throughput include the number of users, MAC layereffects, propagation loss, and multipath and other interfer-

ence sources. A complementary measurement is latency,the time lag for a packet to be transmitted over the

WLAN, which can range from several milliseconds toseconds. Generally, latency, can be highly variable even

under constant conditions. Also, latency and throughput

_ inversely proportional.

]................-::_::,:::::, ................•................................................_:,'::::::, ...........................'::::::::::. ::::::::::::::::::::::::::::::::::::::: _. _._ ..................

iiiIiiiB!,_:';:,::]:=I!II:::_=:":';;:!il::iJ;:i:iiii]i_li:i_ii:..-..-i:ii:?::i-:ii:i:/

iiii.i.--.iiiii iiiii-i"iiiiiii:i!_iI..:i.-"-"!i ii iiiiiIiiiiiiii_l_.-".-"iii_i_iiiili!lli.."..:'..iI_I-"-" _!_)ii:i!_J

:':':';':':'x'-I ":_:-:::-.- •"_-::;.£_" -_':?:?-:"?_:2£"

:+ ..... _ x • i " ".0"-':.'4_ TM::_ __ .... __-.'*!::_: ......

___{!iiiiii!:_........ :'NiI_-g-!i_?_:.: _I": r:_:iil

Figure 2. MEX components

The MEX operational simulations have demonstrated theneed for shorter latencies and more sustainable

bandwidth. To accomplish this, a leading 802.11btechnology based on the Orinoco (formerly LucentWaveLA.N) product was tested and compared to the

Proxim product. The MEX architecture made use of allthe WLAN configurations noted. Peer-to-peer networksare utilized with some of the handheld mobile devices,

with APs at base and EPs fimctioning as repeaters locatedon high hills. Directional antennas are used to increase the

range and connectivi_ of the MEX wireless technologies.

The "unlicensed" nature of radio-based WLANs means

that other products that transmit energy in the same fre-

quency spectrum can interfere with a WLAN system.Microwave ovens are a concern in the 2.4 Ghz range. An-

other concern is co-locating multiple WLANs. "v_ileWLANs from some manufacturers interfere wSth other

WLANs in the same frequency bands, others coexistwithout interference. Often this is an issue of network

domain and channel management practices. Interferenceis measured as a decrease in throughput, range and cover-

age or as an increase in latency and can be confirmed byvarying the interference source or by using a spectrum

analyzer.

Wireless networks can be scaled in two ways: by in-

creasing the number of nodes of a given WLAN or by

layering more WLANs in the same coverage area_ Inter-ference effects limit layering. Scalability can be expressed

by a graph of aggregate throughput versus number ofnodes or by aggregate throughput versus number of

WLAN subnets. Due to complex frequency domain inter-actions, this must be determined using physical testing of

multiple subnets.

A variety of network modeling tools can be applied tohelp simulate performance of candidate architectu.res,

taking into account the complex timing of network data

transfers. Extend, developed by Imagine That. Inc., is an

advancedsimulationtootdesignedto developdynamicmodelsfor realprocessesExtendallowsblocksto beplacedinkierarchicalstructurestobesimplifiedandutil-izedin other parts of the model. For example, transmis-sion delay is a characteristic that is used throughout themodel, and is therefore placed m a hierarchical block for

use in multiple locations throughout the model.

Evenson modeled the WLAN at thc Media Access Con-

trol (MAC) layer of IEEE 802.11 _hat uses the Carrier-Sense Multiple-Access Collision Avoidance scheme. [11]

It specifies that the transmitter must send out a very short

packet called a "request to send" (RTS) before actuallytransmitting the data packet_ Only the destination receivercan respond to the RTS with a "clear to send" (CTS). Allother stations must stop transmitting until the data packetis sent and an acknowledgment (ACK) is received. The

three types of message traffic modeled were the RTS,

CTS, and data packets. The basic component of the Ex-tend MAC model is the modeling of transmission delay.

Transmission delay was modeled by dividing the message

size by the respective link bandwidth to represent the

delay encountered in sending data over the network. TheMEX architecture was modeled using these building

blocks to determine latency variation,;.

One goal of the modeling was to determine the MAC

layer effect on latency. Many iterations of the model wererun to see the effects of changing message size and band-

width. The model would generate plots representing the

adjustment of the network to the traflSc generated by mul-tiple nodes with similar message profiles. As shown in

Figure 3, the latency increases in the beginning of thesimulated 10-second run, as the network is loaded with

traflSc. As time progresses, however, latency varies as thenetwork adjusts to the traffic load coming from eachnode. Due to the functiouing of the 802.11 MAC protocol,the network adjusts to smooth out the effects of initial

latency increases. Notice that there are two sustainablemodes of data transfer in the steady stale, one with lower

average latency than the other.

ii iI•i ' •.•,o.:

Figure 3. Plot of latency varying by

message size over simulation time.

RF PROPAGATION THEORY

A radio system transmits signals through an antenna,

forming a propagating electromagnetic wave that is

received by a complementary antenna and radio.[12] Themaximum distance between a RF transmitter and receiver

depends on several system and path variables includingtransmitter output power, receiver sensitivity and noisethreshold, antenna design, and propagation loss. Terrain

and atmospheric conditions can affect propagation loss.Ideally, isotropic antennas have equal radiation or

reception of power in all directions. Practically, theclosest to this is an onmi-directional dipole antenna,

which spreads the majority of radiated power in a 360-

degree doughnut-shaped belt known as a "main lobe."Directional antennas shape or focus the main lobe energy

in such a way that the majority of it is radiated in a

specific solid angle. "Antenna gain" is a measure of thedirectivity of an antenna. It is defined as the ratio of theradiation intensity in a given direction to the radiation

intensity that would be obtained if the same power wereto be radiated isotropically. The "radiation pattern" of an

antenna is a graphical representation in either polar orrectangular coordinates of the spatial energy distributionof an antenna. "Antenna beamwidth" is defined by the

angle that subtends the two half-power points (-3 dB) oneither side of a directional antenna's main lobe radiation

pattern.

A concern of WLAN design is the loss the carrier signal

will suffer as it travels to the receiving station. Study ofthis consideration is called "link analysis." Immediate loss

is due to the attenuation or "spreading loss" affecting any

transmitted signal, also known as "free space path loss"

(Lfs). By estimating Lfs when designing the link, radiounit capabilities can be estimated to ensure a certainminimum signal threshold for the receiver. Lfs is afunction of the square of the distance from the transmitter,and is shown in the following equation.

where d = distance, )_= wavelength

Radio waves suffer intermittent additional losses in signal

strength through absorption as they travel through the at-mosphere. These intermittent losses, the "fade margin" in

link analysis, may account for additional signal degrada-tion. The total loss for a link can be better estimated once

fxee space and fade margin losses are accounted for. (In-cidental losses in cables and connectors can be addressed

as necessary.) Total path loss is then used to estimate

power received (Pr) at a certain distance. This knowledgewill enable engineers to determine how radiation re-

quirements, antenna specifications and placement, andcell coverage will affect the prescribed coverage and per-

formance objectives.

RFlink analysisdeterminesthedifferencebetweenthenominaloutputpowerof atransmilXerandtheminimuminputpowerrequiredbyareceiver.Foralinktoprovideacertainperformanceobjective,thetransmittedpowerandsystemgainsmustbegreaterthanall thelossesthatim-pingeonthecardersignalasit propagatesthedistance,andthoselossesresidentin theendsystem'sinternalcir-cuitry.Thisbalancingofgainsandlossesisoftenreferredtoasa "linkbudget."LinkbudgetspredicttherehabiliLyof a communicationpathgivenp,trticularsystempar-ameters.A link budget can be expressed by the Friis

equation:

]9r(_) = Pt + Gt - Lf_ - L x + G r [13]

where _P_= power at the receiver (dBm)

Pt = transmitter output power (dBm)

Lfs = flee space path los_ (dB)

L_ = extra losses (e.g. multipath ) (dB)

Gt = Wansmit antenna gain (dB)

Gr = receive antenna gain (dB)

Figure 4 shows the basic set up for network testing.Communications testing consisted of placing the access

point (AP) at a fixed base location and nmning a series ofdirected site surveys to the mobile node EP mounted onthe ATV Several runs of laten_ and throughput data

were collected, and then the distance to the AP was doub-led. From this, range calibration curves were generatedfor various radio and antenna combinations.

Our Ames Research Center test site provided a total dis-

tance of 1.1 km clear line-of-sight with no obstructions,no obvious sources of multipath or interference, and novariations in elevation. The antennas were mounted 6 feet

above the ground at the same exact level at each end. Ex-traneous sources of RF interference had to be eliminated,

and to insure this a spectrum analyzer was used to meas-ure RF energy in the 2.4 Ghz band. The spectrum ana-

lyzer, which delivers a reasonable graph of relative RFenergy, was a sampling Lype provided as a utility, with the

Proxim cards. Test points were started at 10 m and dou-bled in distance at each test point up to the mazdmum1.1 km distance.

M_EX TESTBED DATA

The Proxim system field tests were focnsed on determin-

ing calibration curves for latency and throughput versusdistance using different combinations of omni-directional

and directional antennas. The antennas' had varying per-formance characteristics included beamwidth and gain.

The goal was to find the optimal ante_ma combination forthe intended operational enviromnent. Range andcoverage areas for each system will help define furtherfield simulations.

Procedures were developed based on the diagnostic toolsavailable from the radio manufacturers and methods dev-

eloped by the IMT lab.[14] The Proxim WLAN compo-nents have internal radio diagnostic tools to test parame-

.ters such as throughput and latency through a series ofbroadcast and directed site surveys. In broadcast site sur-

veys, the active nodes of the network were discovered bysending out packets to all nodes within antenna range, andwaiting for responses from all nodes t/tat receive the sig-nal. Once the network topology was established and

proper operation of the network verified, performance in-formation on individuaI links was found through directed

site surveys between individual nodes, based on the MACaddress.

Each antenna combination simulated a "backbone" link of

the MEX architecture such as the primary repeater to mo-

bile ATV link. The following antenna combinations were

tested for the Proxim system during this research:

• 0 dB omm-directional to 0 dB omni-directional

• 11 dB planar sectoral to 0 dB omni-directional

• 24 dB directional parabolic to 0 dB omni-direcfionai

The Proxim radios used for these tests were of the Open-

Air XR series which provides 500 mW output power.They are FHSS types with a raw data rate of 1.6 Mbps.The Orinoco radios provide 32 mW output power and are

802.11b DSSS type with a raw data rate of 11 Mbps. Theradio automatically controls transmission power and

speed. Both radios fall back to lower data rates if linkmargins deteriorate. What are the performance tradeoffsbetween these two different systems measured in the fieldunder ideal conditions and how does this relate to RF link

analysis?

I / [ 1 I11I I I I I II

i--÷4 ............ ....... IlL....... 41.... 4-t-bllIIl • m I r i / 1i-

I l I I I 11m. |: • m: M

ATV base

Figure 4. MEX field test setup

FortheOrinocosystem,high-levelnetworkthroughputmeasurementsweremadeusinglargefile transfers(3MB) throughtheWLAN.Application-levelsustainablenetworkthroughputwasdeterminedusingaTCP/IPfiletransferviaFTP:anFTP server was set up at the base AP,

the mobile ATV computer requested a file from the server,and the file was transferred. File sb'e and time to down-

load were recorded, and average sustainable network

throughput obtained by dividing _e size by time. The testwas then reversed and the same file uploaded to the

server. TNs is done because the choice of antennas and

radio characteristics can result in an asymmetrical con-

nection.

The Orinoco by Agere Systems product was tested with

only the built-in antennas because there were no appropri-ate external antenna connections for it at the time of the

tests. Test tools provided by Agere systems give RF

domain signal levels and network throughput data simul-

taneously. The antenna is a slot dipole with about 0 dbgain_ It is notable that that the built-in antenna has a

unique dispersion pattern resembling a horizontal flathalf-torus. This was experimentally determined by plac-

ing the radios at a distance and measafing the radio-leveldata collected while changing the orientation of oneantenna in relation to the other. FTP throughput and RF

signal level data were collected in the same manner as the

Proxim system.

Finally, formal link analysis of the two systems' RF char-acteristics was done using manufacturers' specifications.

The expected range in a vacuum was compared with the

actual range achieved in air (with water vapor) and addi-tional factors such as atmospheric absorption and parasitic

losses calculated.

The performance of future wireless network designs canbe predicted using this analysis method, and the testbeddata was essential for determining these correction con-stants in the link analysis. The first important results are

800.00

700.00'

600.00"

400.00"300.00

200.00

100.00

the Proxim calibration charts for throughput and latencyestablished for the individual antenna combinations.

Figure 5. shows throughput of data on tested link in Kbits

per second. The generally decreasing throughput atincreasing distances on each test set was expected and inaccordance with theory. Also, the higher the antenna gain

the higher the throughput and the larger the effective dis-tance for communication. These calibration curves are the

result of numerous data runs and represent averages fromseveral runs. However, there is an unexpected dip at mid-

range, when throughput varied about a factor of two. Atthe closest test distance of 10 m, there was highest

throughput and lowest latency, as expected, followed by a

significant drop in throughput at 20 rn. There is a corre-sponding increase in latency (shown in Figure 6) that may

explain the throughput loss. This dip occurs at too short adistance to be the result of flee space losses and low sig-

nal strength. It should be noted that the RF signal strengthbetween the two radios varies significantly with each an-

tenna combination, and yet this dip occurs at nearly thesame distance for each graph. Could this be a MAC layer

timing interaction?

Figure 6 shows average latency values for the series ofpackets sent during the directed site survey at eachdistance. Latency and throughput are inversely related,and this is evident in the data. To test the latency, the

computer sends packets for ten seconds or other selectedinterval and monitors the average, maximum and standard

deviation of the latency of the packets. The packets aresent at maximum rate and therefore also provide the

throughput numbers. This is all measured using the radiodiagnostic tools that come with the Proxim product. The

latency increase and throughput decrease at 20 m has apossible explanatiorL The OpenAir protocol was designed

based on expected use in a commercial environment,typically inside a building. Remembering the RTS/CTSconstruct previously discussed, it is possible that the

protocol simply does not fimction optimally at a range of20 m.

Proxim Throughput

0 56o 10bo 1500Distance (m)

Figure 5. Proxim throughput calibration curves.

..-o.-- 0 dB+11 dB

-'X--- 24 dB

O

120.00

100.00

80.00

60.00

40.00_i

20.000

Proxim Avg Latency

sc;o lo'ooDistance (m)

Figure 6. £roxim latency calibration curves

--_--0 dB"-&- 11 dB--X--24 dB

15OO

The interaction at the MAC layer could have been

optimized for use within an average office space which israrely larger than 10 m square The Extend MAC layermodel showed considerable variations in throughput earlyin the simulation and at least two modes of sustainable

throughput. These variations cannot be the result of RFsigna/ strength variations, but may be the result of

interactions of the MAC layer coinciding with thevariation of free-space propagation times. This is only ahypothesis, but simple modeling of the MAC layer did

demonstrate that it is possible.

After this initial variance, throughput and latency werewell behaved. Throughput increases significantly beyond

the initial dip, and this peak scales with antenna gain.There is a large "sweet spot" plateau where throughput

decreases and latency increases in a manner consistentwith free space propagation losses. The range for eachcurve is proportional to the RF link gain provided by the

antennas, ranging from 0 dBi for the two dipoles to +24dBi for the parabolic antenna. The "sweet spot" seems tobe present in all combinations from abc,ut 80 m to 600 m,

where good link qua/i V is maintained with a relatively

high throughput and low latency. This testbed data can beused to determine actual propagation losses.

The Orinoco DSSS radio system throughput graph fol-

lows (Figure 7). As expected, the Orinoco outperformedthe Proxim in the network throughput test with five times

the throughput, corresponding to the ratio of the raw datarates. This does not come v,Sthout potential drawbacks,

though. A very interesting result is a sharp increase fxom150 meters to 380 meters, evocative of the Proxim result!

The two radio systems are completely different in modu-lation and basic characteristics, and yet there is a similar

variance in throughput versus distance at 20-200 m Both802.11 and 802.11b use similar MAC layer negotiation

protocols. This is yet another clue that this inconsistencyis due to timing interactions in the MAC layer.Subsequently, there is a linear decrease in throughput asdistance is increased to the maximum 1.1 kin_ The Ori-

noco product had significantly greater range than the

Proxim product for reasons quantified in the next section.

Orinoco FTP Throughput

5000

4500

4000"

300O

2500

2000

1500'

1000"

500"

0 260 460 660 800 1000

Distance (m)

Figure 7. Orinoco FTP throughput

1200

RF LINK ANALYSIS

Conventional RF link analysis was applied to both radio

systems using the manufacturer's specifications in orderto determine theoretical operating range in a vacuum.Transmitter power, antenna gain and receiver sensitivityare used to calculate maximum range considering only

free space loss. Of course, in the real world one mustconsider atmospheric effects and parasitic losses as well.Since these extra losses are unknown initially, one can

work backward, comparing the theoretical distances tothose measured in the field to detemfine additional losses

in the field test setup. These additional losses can befarther analyzed to determine if there is a component that

scales linearly with distance and which would therefore

represent the effect of atmospheric water vapor. Thefrequencies used for these radios, 2.4 GHz, lie within thewater absorption band and therefore water vapor will

affect propagation of the signals on earth,

The following table demonstrates quite dramatically thatthe Proxim radio systems have significantly less range

than predicted from free space losses alone. A major

initial impetus for these tests is that the MEX fieldsystems did not perform as well as expected anddete _rmming the cause could improve performance.However, the amount of additional loss for the Proxirn

system is rather high_ The "link budgef' represents themaximum fade margin in dBm, and the "max range"represents the corresponding theoretical operating limit invacuum. "Actual range" is determined by analyzing

throughput versus distance graphs for each radio system.

Determining "actual range" requires _mderstanding radio

operation details to correctly interpret field data, and thisinterpretation is somewhat different for the two radiosystems Many years of wireless net_vork testing haveresulted in a clear understanding of the behavior of the

serf-adapting modulation feature of these radios. It falls

back from quad phase shift keying to bi-phase (BPSK)

keying when the radio signal is too weak to support flailrate dam transfer. The Proxim system switches from a 1.6

Mbps rate to 800 Mbps just before a loss of connectionoccurs. Estimating the point at which these rate changesoccur in the field tests allows estimating the actual range.

The Proxim system has a 1.6 Mbps raw data rate which

actually delivers less than half that (typically 33%) insustainable throughput due to packet-- and MAC-laveroverhead. Therefore the initial slope of throughput falloffis used to estimate the point at which throughput falls off

below 200 Kbps. This is the point at which BPSK is no

longer effective, being 25% of the raw data rate of 800

Kbps which is applied consistently to the data sets. Thisactual range is then used to calculate the additional lossdue to all effects in the field tests. These "additional

losses" (Addit'l) are very high, ranging from 16-33 riB.

Further analysis attempted to identify parasitic losses andcalculate atmospheric losses. Cable and connector losseswere calculated from manufacturer's specifications to

determine total "cable loss" for each test setup. Parasiticcable losses can be calculated readily, so there cannot be

any unaccounted-for losses in the system except from at-mospheric absorption. The use of long cables (36 ft) toconnect the Proxim AP to the 11 and 24 dBi antennas (but

not 0 dBi dipole) added a significant extra loss of 2.6 dB,

reducing output power by nearly one-half Furthermorethe Proxim EP was connected to the ATV antenna by a

six-foot length of RG-58 losing another 2 dB total and thecalculation takes this into account. The remaining loss

was divided by the actual test distance to determine addi-tional attenuation per meter ("atmospheric loss" [Atmo]).This is about 0.02 dB per meter, comparable to ultra-lowloss cable. Propagation of 2.4 GHz FHSS signals through

the atmosphere results in significant attenuation primarily

due to absorption in water vapor.

Antenna

0 dBi dipole

11 dBi sect.

24 dBi parab.

XmitPwr

(dBm)

27

27

27

i;_ecvSens

dBm]

85

85

85

Table i. Proxim RF link analysis

Proxim XR

Ant Link Max Actual

Gain Budget Range Range(dBi) (dB) (m) (m)

0 112 3879 600

11 123 13764 900

24 136 61482 1350

Addit'lLoss(dB)

16.2

23.7

33.2

CableLoss

(da)

2.0

4.6

4.6

AtmoLossdB/m

0.0237

0.0212

10212

The Orinoco system was tested with only one antenna, so

the RF analysis was performed on only one system

configuration but in three differen_ bandwidth regimes.The Orinoco radio has a much lower receiver sensitivitythan the Proxim, 6 dl3 more sensitive at 2 Mbps. Agere

provides receiver sensiti-,ity specifications tied to rawdata rate, and the Orinoco kas four different data rates: 11

Mbps, 5.5 Mbps, 2 Mbps and 1 Mbps. Using similar

assumptions, one finds that only 11 Mops raw rate cansupport sustainable throughput in excess of 2.8 Mbps,

only 5.5 Mbps can sustain over 1.6 Mbps and only 2Mops can sustain over 0.6 Mops. Cable loss wasminimized since the antennas were directly attached to the

radio PC-cards. Determining actual range for each rawdata rate and sensitivity, we can calculate the additional

loss at a nearly constant 4.5 dB, The atmospheric loss ofthe Orinoco at I1 Mbps is haft that of the Proxim system

as shown in Table 2. Further armlysis of the threebandwidth regimes also provides an intriguing result, that

the atmospheric attenuation in each band decreases as the

bandwidth decreases. Of course, this is intuitively correct,

since it takes a better signal to noise ratio to correctly

interpret modulation at a higher data rate. This first order

approximation of linear propagation loss does not accountfor dispersion and other effects, but should be useful forestimating new WLAN performance characteristics.

Finalt3; the radio diagnostic tools provided by Agere forthe Orinoco were used to create a graph of received signallevel versus distance as shown m Figure 8 below.J15] At100 m the level is about -60 dBm and at 1000 m it is

about-87 dBm. The difference is 27 dB, which corre-

sponds with 20 dB due to free-space loss, for a 10 to 1distance change plus the 5 dB additional loss calculated

above. This graph (Figure 8) is constructed directly fromthe radio circuits of the Orinoco calibrated in d]3m and

acts as an important cross-check for the RF linkcalculations.

Bandwidth

11 Mbps

5.5 Mbps

2 Mbps

XmitPwr

(dBm)

15

15

!5

R{.nCV

Sens

(dBm]

3

7

c!

Table 2. Orinoco RF link analysis

Orinoco 0 dBi antennas

Ant Link Max Actual Addit'l

Gain Budge1 Range Range Loss(dBi) (dB) (m) (m) (dB)

0 98 774 450 4.7

0 102 1227 750 4.3

0 106 1944 1200 4.2

CableLoss

(dB)

0.2

0.2

0.2

AtmoLoss

(dB/m)

0.0100

0.0054

0.0033

0.00 I-lo.oo i-2o.oo/-30.O0,L

m -, 0.00 L"o -50.00

-60.00

-80.00"-90.00"

-100.000

Orinoco RF Signal Level

I i | I

200 400 600 800Distance (m)

Figure 8. Orinoco Radio Signal Levels

i

1000

INTERFERENCE AND SC,_LABILITY

Interference caused by electroma_metic energy in thesame frequency band used by WLANs to communicate

data can cause a loss of throughput, increased latency andeven loss of connection. The same metrics used to char-

acterize performance can be used to study the effects of

interference. Furthermore, interference effects primarilylimit scaling up WLANs by using multiple co-located

WLAN systems to increase aggregate bandwidth. We

study the effects of interference on the two WLAN sys-tems and attempt to layer multiple WLANs to determine

scalability limits.

The interference tests were desit_ed to stress the

robustness of the data link in each radio system when

subjected to a strong in-band interference source. Thesource used to create the interference is an analog videotransmission device (Trango) with four selectablechannels all within the 2.4 GHz ISM band providing

fairly high average and peak signal strengths. This source

was placed between two WLAN node, s spaced about 3 mapart_ The interference source was placed both within andoutside the current working bandwsdth of the WLAN

system and the effect on link per_0rmance measured.Figure 9 shows the RF spectrum of the interferencesource, mostly in the lower part of the ISM band, obtained

using the Proxim samplLrig spectrum analyzer.

The Proxim and Orinoco systems produce very different

spectrums, as shown in Figure 10. Notice the differencebetween the peak (gray) and average values (black) ofeach bar. Each bar represents a 1 MHz slice of the 2.40 -2.483 GHz ISM band -- 83 bars total. The Proxim

produces narrow, 1 MHz-wide signals with high peak and

average energies representing the frequency hops. AnOrinoco rtmning on Channel 4 produces a broad spectrumof much lower peak and average energy, pu_ng its

primary information into the small hump of average

power at the lower end of the specmam. The DSSS radiospreads its information across a larger part of the ISM

band, providing higher raw data rates compared to FHSS,by pu_ing the information into more simultaneous signalbands. The observant reader will notice an FI-ISS hop in

the Orinoco spectrum and residual DSSS energy in the

Proxim spectrum since both systems operate simultane-

ously in our test lab.

The interference source puts significant peak and average

energy into a very limited portion of the band, which caninterfere with WLAN signals in the same area. If an inter-

ference source is placed within the current operating bandof the Orinoco, the entire WLAN link is lost and

throughput is zero. If the interference source and WLANchannel allocation are shifted apart, then there is minimal

throughput loss. The Proxim system, however does notlose the link and only about 10% of its throughput whenan interference source is present, regardless of channel

settings. As expected, the higher bandwidth DSSS systemis much more sensitive to in-band interference.

_ ....... _iiiiiiiiiiiiiiii:_)i2ii?i! i!........................................................ Signal S_rer'_th _ Freqt._ncy ........................................................

2.402 Frequencies [GHzl 2.480

;"Le9 er_d............_:;&.-:: , _?:i:i: i

: Peak Hok_ :_:::: .--3amen{} Rate .................................... .

F:gure 9. Interference source spectrum

S

g

t3

S

t

I

n

g

,ill

Z40Z

i PeakHold

;hOOp

.................................................... Signal _uency .....................

[I =I i .IL i i x i

Fr equencie,_: [GHz) Z4802.402 :- Legend ..................

Peak Hold _l_++:i .-Sarr'+p_]r_gR at e ........... _.......................

D[_I 2 : Average I i';'_ _ I_ ed-=.+m i"" Fast Help t

Figure 10. Orinoco Channel 4 spectrum (top) and Proxim spectrum

"Scalability" is defined as how man3' WLAN nodes or

radio systems can be nm in the same v_cinity. Interferenceis the prime factor in salability by reducing connectionreliability and by reducing throughput. Keeping interfer-

ence susceptibility to a minimum allows for maximumscalability. We initially considered the Orinoco and

Proxim systems individually. Proxim has developed fre-quency'hopping radios with fifteen different hopping se-

quences. Theoretically, fifteen separate networks couldoperate at the same time, providing up to 15 times the ag-

gregate bandwidth of a single subnet. In reality this num-ber is only five, and after five networks have been set upin the same vicini_, the performance of all decreases

sharply. This was tested in much the same way as the in-

terference tests, simply running five Proxim subnets in

our lab, with each node about 3 m from another within the

10 by 15 m lab.

The Orinoco system can only support three co-locatednetworks. Orinoco recommends that channels one, seven,and eleven be used to minimize channel overlap, which

can be confirmed by looking at the Orinoco channel

structure on the spectrum analyzer. This was also con-firmed by tests in the lab. Notably, adding an Orinoconetwork does not have much affect on the Proxim system,

given its much greater immunity to interference. Con-versely, a Proxim system has minimal effect on an Ori-

nocosystemsincetheProximspect]umdoesnothavethehighaveragevaluesoftheinterferencesourceusedprevi-ously.Ourlabsupportsbothsystemsrunningsimultane-ouslywithoutobservinganynegatiw,•interactions.

CONCLUSION_

Theseexperimentsprovidedtheempiricaldatarequiredfor beginaingthedesignof a secendgenerationMEXcommunicationarchitecture.Theexperimentaltoolsandprocedureshavebeenverifiedovera periodof manyyearsinavariety of environments; however, tiffs was thefirst time outdoor testing was condu_Xed under controlled

conditions. Operationally, the MEX required more robustnetwork links and higher throughput than the original

Proxim system provided. These tests yielded data on whateffects caused the links to perform more poorly than

expected: high losses in cables and connectors, variablelatency and throughput in the mid-range, and high

atmospheric absorption.

Although cables were constructed according to accepted

practice using low-loss cable suitable for the frequencyrange, losses at 2.4 GHz are double those for 1 GHz fre-

quencies. This large factor .requires significant changes inconventional engineering practices when dealing withWLAN technology. For example, the radio units shouldbe mounted on the repeater mast, since this is the only

way to reduce the cable length and hence losses. This

complicates the mechanical and electrical design of therepeater unit. These issues are par_cularly important forWLANs since they use low-power transmitters whichcannot tolerate additional losses. Furthermore, as WLAN

frequencies and bandwidths increase, these factors will

become even more important.

Other WLAN performance issues :tre related to theserious variations in throughput tha! were observed atmedium range, hypothetically due to MAC layer timinginteractions. These variafons manifested themselves as

poor performance of WLAN links near base camp andcould not be explained orl the basis of RF signal levels,which were high. Rather, the performance variationsresulted from small changes in mobile ATV position and

even as changes in link performance o_er time. This couldnot be explained prior to creating the calibration curves

which showed they are due to unusually high latenciesunder certain mid-range conditions.

A major surprise of the testing was the difference betweenFI-ISS and DSSS radios in regard to atmospheric absorp-tion. The data show a two-to-one difference between the

Proxim and Orinoco at fifll bandwidtta (1.6 Mbps vs 11

Mbps) and an eight-to-one difference at similar band-

widtk This particular effect was too pronounced to be at-tributable to experimental error. The Proxim system cannot cover 1 km without a directional aatenna, unlike the

Orinoco. It is speculated that the higher power levels ofthe FHSS radio cause greater excitation of water mole-

cules and therefore greater absorption and that this effectis not linear. However, the DSSS radio's ability to cover

greater range at greater bandwidth has significant trade-offs. Interference testing clearly demonstrated increased

susceptibility to in-band interference, which translatesinto increased susceptibility to multipath interference

caused by walls, hills or any other obstruction. Therefore,

in hilly terrain, the DSSS may not have a real range ad-

vantage since it is more likely to suffer degradation due to

multipath.

Finally, the two different radio systems scale differently.Five Proxim subnets would deliver 8 Mbps of aggregateraw data bandwidth while three Orinoco subnets would

deliver 33 Mbps. The 2.4 GHz ISM band provides a totalof 83 MHz of bandwidth: neither radio system fills this

completely. There are improvements possible in providinghigher bandwidth. One practical approach is to run FHSSand DSSS systems concurrently, providing a balance ofdesirable characteristics. We will attempt to determine

scalability of mixed mode WLANs in future tests.

Successful design of WLAN systems requires the com-munications engineer to have technical competence in

both the radio and computer network domains -- a chal-

lenging role. Radio analysis is required to provide ade-quate link margins, and network expense is required tounderstand radio subnet setup, MAC layer tuning andoverall architecture. The anal3_cal technique for lumping

all atmospheric effects, including wave dispersion, intoone linear constant is useful for estimating the perform-

ance of new designs. Future work will include design and

test of a hybrid communication system incorporating thebest aspects of both FI-ISS and DSSS systems. MAC layer

timing interactions should be further explored.

REFERENCES

[1] Wireless LAN Association. What is a Wireless L4N?

[http://wwwwirelesslan.com/wireless/] February, 1999.

[2] R. Alena, Wireless Network Experiment- RiskMitigation Experiment 1306, report on STS-74/76-Mir20

experiment for the Phase One International Space Station

Program, 1996.

[3] R. Alena, E. Yaprak, and S. Lamouri, "Modeling AWireless Network For International Space Station," IEEE

Aerospace Conference, 1999.

[4]K.Feher.Wireless Digital Communications."

Modulation and Spread Spectrum Applications. Prentice

Hall, April, 1995.

[5] W.E. Lutz. Wireless LAN Technol,_gies.[http ://licensedinfo. s.nsa/FACCTs/00014337.htm].

January, 2000.

[6] J. Geier, I_?reless LANs." Implementing InteroperableNetworks, (pp. 93-131), Macmillan Technical Publishing,1999.

[7] NASA Human Exploration and D:_'elopment of Space

[http://www.hq.nasa.gov/osf/heds/]. August, 2001.

[8] Stephen Braham, Peter Anderson, Pascal Lee, RichardAlena, and Brian Glass, "Canada and :Mmlogue Sites for

Mars Exploration," Proceedings of the Second Canadian

Space Exploration Workshop, 1999.

[9] Stephen E Braham, Richard Alena_ Bruce Gilbaugh,and Brian Glass, "Space Networking and Protocols for

Planetary Exploration and Analog Plarletary Sites," IEEEAerospace Conference, 2001.

[10] R. Alena, B. GilbaugK and B. Glass,"Communication System Architecture for PlanetaryExploration," IEEE Aerospace Conference, 2001.

Richard Alena is a Computer Engineer and the Group

Lead for the Intelligent Mobile Technologies (IMT) Lab

and the Mobile Exploration System testbed at NASAAmes Research Center. The IMT team integrates mobile

hardware and software components into unique systems

supporting human performance for flight and payload op-erations aboard spacecraft. He was principal investigatorfor the Wireless Network Experiment flown aboard Shut-

tie and MAr, technology later adopted by the International

Space Station Program. Rick spent three summers in theCanadian Arctic developing mobile technologies for hu-man planetary exploration. He has a MSEE&CS from

University of California, Berkeley.

[11] Proxim Corporation RangeLA?vT, TechnicalReference Guide. Pro:din, Inc. USA,

[gttp://wwwproxim. corn], 2000.

............. .:.::i_ill_ :_::':':-_

- .[L2_BreezeCOM_Wireless IMNConcepts: Appendix B .... ::::::::::::::::::::::........... _:::::':::'_:::_

[http://www.Breezecom com/TechSupport/bnpro-ab, htm]. __iNi_

Augu 2000..... ...:. ::_'_!i "-: y:_._

[13] W. romasi. Electronic Communica:ions Systems." ii__Fundamentals Through Advanced. Prenlice Hall, USA, __

1998.

Darin Evenson (below left) is a L!eutenant_ in the United

States Navy and a graduate student at the Naval Post-graduate School (NPS) in Monterey, California. He is in-

terested in applying mobile wireless technologies to smallgroups of people deployed in hostile environments. Hecompleted his tmdergraduate work at the United States

Naval Academy and has recently graduated from NPSwith a MS in Information Technology Management and

an MS in Space Systems Operations.

:::::::::::::::::::::::::::::::::::::::

?..:i;?:-.:..'ii.";ii-iii:

:!:_:!:i_?:i:i:i:!:i:

[14] Bruce L. GilbanglL Brian Glass, and Richard Alena(2001) "Mobile Network Field Testing a_ HMP-2000,

!EEE Aerospace Conference, 2001.

[15] Orinoco User's Manual

Victor Rundquist (above right) spent two years workingas a master electrician for a sound and fighting companyin Phoenix, AZ and later went on to manage a retailelectronics store. Victor has been attending De Anza

College for three years taking courses in advanced

mathematics and the Physical Sciences and will be

graduating in 2005 with a degree in ElectricalEngineering and Computer Science. He has been workingat NASA/Ames Research Center in the Intelligent Mobile

Technologies group for the past year, through theFoothill-De Anza College Internship Program.